Hi,
On 2019-04-02 13:30:56 -0400, Tom Lane wrote:
> But it's hardly plausible that that runs post-commit.
Yea, one surely would hope that that's not happening. I suspect it's
possible it's running pre-commit and somehow the failure handling is
screwed up leading to the 'already committed' error.
> So now I'm thinking that what we really need to know is why a
> TM_Invisible result was returned. That's going to be a bit harder ...
Yea. I re-skimmed through the code and can't immediately see anything
wrong - not that such subtle issues would necessarily be noticed without
more information.
I wonder, is there a chance that you're using an extension with xact
hooks, or a modified version of postgres? Could you show us the exact
schema used, and the config?
Greetings,
Andres Freund